{"id":"21b3ab51-8b72-4556-bdb5-0fdd3ddfb257","arxiv_id":"2608.02182","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Frequency map analysis converges with error O(e^{-cT^ζ}) for analytic quasi-periodic functions and with super-polynomial rates for Brjuno and almost-periodic cases.","lead":"The paper proves that Laskar's frequency map analysis converges exponentially fast when the underlying signal is analytic and the frequency satisfies a Diophantine or Brjuno condition, instead of only at the polynomial rates known before. It also extends this guarantee to almost-periodic signals with very general spatial structure, which matters for precision frequency extraction in celestial mechanics and accelerator physics.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The exponential rate in Theorem 1.2 rests on the unproved L^1 derivative bound for w_{p,q}^{Las} imported from [TL25b, Lemma 4.1]; without an independent proof of (2.8), the central small-divisor estimate in Lemma 2.3 is unsupported.","rationale":"The reader’s verdict and weakest-assumption identification are sound. The paper’s central claim is Theorem 1.2, and its proof funnels through Lemma 2.3, which uses the imported L^1 derivative bound (2.8) to control the small-wave-vector sum after repeated integration by parts. Without that bound, the exponential rate does not follow; with only the qualitative C∞ decay used in Theorem 1.1, one gets only polynomial rates. The derivative bound is self-cited, not restated, and is not machine-checked or independently proven in this manuscript, so it is a genuine conditional risk. I considered other possible objections: the almost-periodic small-divisor estimates omit the k−e_1 shift, and Case (III) of Theorem 1.5 is sketched, but these affect auxiliary theorems and appear repairable by uniform factors. The derivative bound is the one step that literally controls the exponent in the headline theorem. Since the reader already marked the verdict CONDITIONAL for the same reason, no change to the verdict is required. I therefore recommend UNCHANGED: the paper should be accepted only after the derivative bound is verified independently or its proof is included.","tokens_in":27277,"tokens_out":13616,"duration_ms":107098,"concrete_test":"Independently derive an explicit Faà di Bruno bound for ∥D^l w_{p,q}^{Las}∥_{L^1(−1,1)}—at minimum for the one-sided case w(x) = exp(−(1−x)^{−q})—and compare with (2.8). As a numerical corroboration, compute ∥D^l w_{p,q}^{Las}∥_{L^1} for l = 1..30 using high-precision quadrature on (0,1), fit the growth rate to the form λ^l l^{β l}, and check whether β = 1 + 1/q holds to within a few percent. If the fitted exponent exceeds 1 + 1/min{p,q}, re-optimize ι in Lemma 2.3 and recompute the exponent ζ in Theorem 1.2; if the bound fails outright, Lemma 2.3 has no replacement and the exponential-rate claim collapses.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1.2 claims ν_T^1 − ν_1 = O(e^{−c_I T^ζ}) with ζ = (τ+β)^{-1}. The proof of the principal-part estimate, Lemma 2.3, rests entirely on the derivative bound ∥D^l w_{p,q}^{Las}∥_{L^1} ≤ λ^l l^{β l} with β = 1 + 1/min{p,q}, imported as [TL25b, Lemma 4.1]. This bound is not restated, proved, or verified here; footnote 8 only says the weighting functions differ slightly in form and that the analysis remains parallel. The optimization of the integration-by-parts order ι in (2.10)–(2.11) is exactly sensitive to the factorial-like exponent: with β as stated, choosing ι* ≈ c T^ζ cancels the algebraic factor T^{βζι} and leaves exp(−β c T^ζ). If the true exponent were larger—e.g. β + δ, or if an extra power of l appeared—then the optimal exponent would change from ζ to (τ+β+δ)^{-1}, weakening the headline rate; if the bound failed badly, the principal-part sum would be uncontrolled and only the polynomial rate of Theorem 1.1 would remain. The cited lemma is plausible (for w(x) ~ exp(−(1−x)^{−q}), derivatives grow roughly like (const)^l l^{(1+1/q)l}), but plausibility is not a proof, and the paper provides no independent evidence that (2.8) holds uniformly in the form used. Because Theorem 1.2 is the paper’s foundational claim and Theorems 1.3–1.5 use the same mechanism, this is the most load-bearing unsupported step.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper revisits Laskar's frequency map analysis (FMA) and claims exponential convergence for analytic quasi-periodic and almost-periodic signals when the parameterized Laskar weighting function w_{p,q}^{Las} is used. Theorem 1.2 states that, under a Diophantine condition with exponent τ, the first-frequency error satisfies |ν_T^1 − ν_1| = O(e^{−c_I T^ζ}) with ζ = (τ + β)^{-1} and β = 1 + 1/min{p,q}, improving the classical polynomial bound of Theorem 1.1. Theorem 1.3 gives an exp(−c log T / log log T) rate under a Brjuno condition. Theorems 1.4 and 1.5 extend the result to analytic almost-periodic functions on Bourgain-type and general lattices, with rates depending on the spatial weight Φ. The proofs split the Fourier index set into a small-divisor principal part and a tail, then optimize the number of integrations by parts after applying Laskar's transform to the weighted window.","tokens_in":27796,"tokens_out":7985,"duration_ms":65748,"significance":"If the main theorems are correct, the paper gives the first exponential convergence rates for Laskar's frequency map analysis, with explicit dependence on the arithmetic of the frequency vector, the analyticity radius, the anisotropy of the spatial lattice, and the parameters p, q of the weighting function. The two-scale decomposition and the optimization of the integration-by-parts order are natural and the main rate in Theorem 1.2 is sharp-looking: the exponent ζ = (τ+β)^{-1} arises from exactly balancing the Diophantine denominator against the factorial-type derivative growth. The paper is a theoretical contribution and does not contain numerical experiments, code, or machine-checked proofs. Its central claims are falsifiable in the sense that explicit rates are stated. However, as detailed below, several load-bearing technical inputs are imported from the authors' previous papers or only sketched, so the manuscript needs substantial revision before the claims are fully supported.","major_comments":[{"comment":"The principal-part estimate rests on the L^1 derivative bound ∥D^l w∥_{L^1} ≤ λ^l l^{β l} with β = 1 + 1/min{p,q}, imported from [TL25b, Lemma 4.1]. Footnote 8 only says the weight functions 'differ slightly in form' and that the analysis is 'parallel.' This bound is load-bearing: the chosen order ι* and the final exponent ζ depend precisely on β. The manuscript should either state and prove the bound for the actual w_{p,q}^{Las}, or give an exact statement with the constants and the adaptation spelled out. Without that, Lemma 2.3 and hence Theorems 1.2–1.5 are unsupported at their central point.","section":"§2.1, Eq. (2.8) and Lemma 2.3"},{"comment":"The small-divisor estimates are applied to k rather than to k − e_1. Since Ω_k = ⟨k,ν⟩ − ν_1 = ⟨k − e_1,ν⟩, the infinite-dimensional Diophantine condition (1.6) must be applied to k − e_1. This matters already for k = 0 (which belongs to Θ) and for k with zero first component, where k − e_1 has one extra non-zero component. In the quasi-periodic proof this shift is handled explicitly via k* = k − (1,0,...,0); the almost-periodic proofs do not do the analogous step. The omission is probably repairable by absorbing an extra constant in the denominator, but as written the displayed lower bounds in Lemmas 2.7 and 2.9 do not follow from (1.6).","section":"§2.3, Lemma 2.7 and §2.4, Lemma 2.9"},{"comment":"The summability argument uses the cardinality estimate #{k∈Θ : |k|_η=ϑ} ≲ ϑ^{ϑ^{1/η}}, imported from [TL24b, Lemma 8.4]. This is a second unproved technical input in a chain of new claims. The manuscript should state this estimate as a lemma and either prove it or give a precise reference with the exact hypotheses. In addition, the sentence 'for any constant c_{22} > 0' is not correct: from |a_k| ≤ C_f e^{−r|k|_η} one can dominate by e^{−c_{22}|k|_η} only for c_{22} ≤ r. The later use in Lemma 2.8 requires c_{22} < r, so the quantification should be fixed.","section":"§2.3, Eq. (2.22)"},{"comment":"The proof of Case (III) is only a verbal construction: one is told to choose K(T) large and Γ(K(T)) small, and then 'such a Φ(x)' exists. No rigorous construction is given for a single increasing Φ satisfying the standing assumptions, Λ(x)=x, the truncation equation (2.30), and the two inequalities displayed before the final estimate. Since Case (III) asserts that arbitrarily fast sub-exponential rates I(T) with I(x)=o(x^β) are attainable, this requires a real existence proof, not a heuristic. Please provide a concrete family of Φ or a fixed-point/selection argument.","section":"§2.4, Case (III) of Theorem 1.5"},{"comment":"The theorem states a rate for any ρ < 1 + η, but the proof in Lemma 2.7 uses the decomposition with |k|_η ≤ (log T)^ρ and explicitly requires 2 ≤ ρ < 1 + η. If ρ < 2 is intended, an additional argument is needed; if not, the statement should be restricted to ρ ∈ [2, 1+η). The same issue affects the final sentence of the proof, where '2 ≤ ρ < 1 + η can be chosen arbitrarily' is not equivalent to the stated quantifier.","section":"Theorem 1.4 statement vs. proof"}],"minor_comments":[{"comment":"The transition from (2.9) to (2.10) drops the factor 2 coming from the Leibniz-rule estimate. This is harmless if c_2 is redefined to absorb it, but as written the inequality has an extra factor 2.","section":"§2.1, Eq. (2.10)"},{"comment":"The symbol ν_1 is used both for the first scalar frequency in the leading term e^{iν_1t} and as the first component of the frequency vector ν in the Diophantine condition. The distinction is clear from context but deserves a sentence, especially because the almost-periodic setting uses ν_1 in both senses as well.","section":"§1, notation"},{"comment":"The use of the implicit function theorem at the point (0,+∞) in the compactified domain is not standard. A short rescaling argument, for example setting s=1/T and applying the usual IFT on a neighborhood of (0,0), would make the step rigorous and easier to follow.","section":"§2.1, Lemma 2.2"},{"comment":"The sentence about the weighting functions differing slightly should be expanded. Since this difference is the only justification for transferring [TL25b, Lemma 4.1], the reader needs to know exactly how w_{p,q}^{Las} relates to the weight analyzed there.","section":"§1.2, footnote 8"},{"comment":"The final constant c_II is said to be 'arbitrarily large.' This is a qualitative statement; it would be clearer to write that for every prescribed C>0 the bound holds with c_II ≥ C for T large enough.","section":"§2.2, Theorem 1.3"}],"recommendation":"major_revision","confidential_remarks":"The paper has a large number of references to the authors' own prior work, and the two most technically delicate inputs (the derivative bound for w_{p,q}^{Las} and the lattice cardinality estimate) are taken from those papers without proofs. For a journal submission, importing published lemmas can be acceptable, but here the central rate claims stand or fall on these lemmas and on the almost-periodic shift issue, so the current manuscript is not yet self-contained enough to certify the advertised theorems. The overall direction is promising and the finite-dimensional Diophantine proof is structurally coherent, so I see this as a major-revision rather than a rejection, provided the missing technical steps are supplied or precisely located."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real result, not a repackaging. Theorem 1.2 gives exp(-c T^zeta) for Laskar's frequency map analysis under analyticity and a Diophantine condition, upgrading Laskar's classical polynomial bound, with zeta = (tau + beta)^{-1} depending on the Diophantine exponent and the window parameters p,q. The main line of proof is coherent: the argmax reduction, the implicit function theorem step, the split into small and large wave vectors, and the integration-by-parts optimization all check out. The Brjuno and almost-periodic extensions are plausible, and Theorem 1.5 is a reasonable abstract framework even if its Case (III) is only a sketch.\n\nThe soft spot is exactly the one the stress-test flags. Lemma 2.3, which controls the principal-part sum, rests on the bound ||D^l w_{p,q}^{Las}||_{L^1} <= lambda^l l^{beta l}, imported from [TL25b, Lemma 4.1] without proof or full restatement. Footnote 8 says the weighting functions differ slightly in form and the analysis is parallel. That lemma is load-bearing: it sets beta, hence zeta. If the true exponent were beta + delta, the headline rate would weaken; if the bound failed altogether, the exponential rate would collapse to Laskar's polynomial bound. The estimate is plausible — I'd expect derivative growth of roughly this order for the double-exponential window — but plausibility is not proof, and the paper's central claim depends on it. A referee needs to see the lemma stated precisely for this window and proved.\n\nThere is also a small gap in Theorem 1.4: the small-divisor argument works with k instead of k - e_1, omitting the shift in the infinite-dimensional product. I think the same estimates can absorb the shift — it only adds a logarithmic factor — but as written it is a genuine omission. The general lattice proof also imports a cardinality estimate from [TL24b] without proof; minor, but worth restating.\n\nOn citations: heavy self-citation, but mostly to the authors' own weighted Birkhoff average papers where the techniques actually come from. That is legitimate, not circular. No fitted parameters, no data, no machine-checked proofs; this is a theory paper, and the practical claims are asymptotic.\n\nWho this is for: people doing FMA/NAFF in celestial mechanics and accelerator physics who want theoretical guarantees, and the KAM/weighted-Birkhoff community. It deserves a serious referee. My recommendation: send it out, ask the referee to verify (2.8), fix the shift in Theorem 1.4, and expand Case (III) of Theorem 1.5 before final acceptance.","headline":"The finite-dimensional exponential bound for FMA is likely correct and genuinely new, but the proof's core relies on an unproved imported derivative estimate; a referee should verify that before acceptance.","tokens_in":28198,"tokens_out":2897,"would_cite":true,"duration_ms":40074,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37J40","70H12","70H08","37K55","70-08","37M99"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that frequency map analysis recovers fundamental frequencies from analytic quasi-periodic signals with an error that decays exponentially in the window length, given a Diophantine frequency vector.","keywords":["frequency map analysis","exponential convergence","quasi-periodic functions","Diophantine condition","Brjuno condition","almost periodic functions","weighted Birkhoff averages","weighting functions"],"falsifier":"Take a concrete analytic quasi-periodic function with a known Diophantine frequency vector, compute the frequency map estimate nu_T^1 numerically for increasing window lengths T, and check whether |nu_T^1 - nu_1| decays like exp(-c T^zeta) for a range of positive c; alternatively, directly test the bound ||D^l w_{p,q}||_{L^1} <= lambda^l l^{beta l} for large l by symbolic or numerical differentiation, since one counterexample would disable Lemma 2.3.","tokens_in":27149,"feed_emoji":"🎯","tokens_out":5224,"duration_ms":43704,"temperature":0.7,"pith_summary":"Frequency map analysis is a standard numerical method for extracting oscillation frequencies from short time windows, originally shown to converge only polynomially. The paper claims that for analytic quasi-periodic signals whose frequency vector satisfies a Diophantine condition, the recovered frequency error drops like exp(-c T^zeta), where zeta depends on the Diophantine exponent and on the shape of the weighting function. This would strictly improve the classical polynomial bounds, making the method dramatically more accurate on long windows. The same strategy is extended to Brjuno nonresonance and to analytic almost-periodic signals with infinite frequency sets under weighted-lattice spatial structures.","feed_headline":"Frequency map analysis errors now proven exponentially small","feed_subtitle":"For analytic quasi-periodic signals with Diophantine frequencies, the recovered frequency error becomes exp(-c T^zeta) rather than a power l","key_machinery":"The load-bearing object is the generalized exponential weighting function w_{p,q}(x) = c_2 exp(-(1+x)^{-p}(1-x)^{-q}), a smooth filter that vanishes flat at the boundaries. Its relevant property is a bound on the L^1 norms of high derivatives: ||D^l w_{p,q}||_{L^1} <= lambda^l l^{beta l}, with beta = 1 + 1/min{p,q}. This derivative control lets the proof optimize the number of integration-by-parts steps as approximately l_* ~ T^zeta, turning the polynomial small-divisor factors into exponential decay. The index split ||k|| <= T^zeta separates the principal part, controlled by the flatness of the weight and the Diophantine condition, from the remainder, controlled solely by analyticity.","core_discovery":"The central claim is Theorem 1.2: for an analytic quasi-periodic function f(t) = e^{i nu_1 t} + sum_k a_k e^{i<k,nu>t} with Diophantine frequency vector nu, and for the generalized exponential weighting function w_{p,q}(x) = c_2 exp(-(1+x)^{-p}(1-x)^{-q}), the frequency map analysis error satisfies nu_T^1 - nu_1 = O(e^{-c I T^zeta}) with zeta = (tau + beta)^{-1}, where beta = 1 + 1/min{p,q}. The proof splits the Fourier index set at ||k|| <= T^zeta. For the small-index part it applies repeated integration by parts to the weighting function, using both the Diophantine lower bound on frequency mismatches and a claimed bound on high derivatives of w_{p,q}; for the large-index part it relies on","pith_inferences":["The proof rests on an auxiliary derivative bound for w_{p,q} that is imported without proof from a companion paper; verifying this bound directly for specific p and q would be the fastest way to test whether the exponential claims are fully supported.","The paper proves upper bounds but not optimality; numerical tests on a known two-frequency analytic signal could reveal whether the predicted zeta is sharp or whether a different exponent governs the actual error decay in practice.","The general-lattice result suggests a design principle: one can choose the weighting function to match the spatial structure of the frequency set, potentially yielding application-specific filters with customized convergence rates.","Since the paper is purely theoretical, a natural testable extension is to compare the empirical frequency error against the predicted exp(-c T^zeta) rate for a few simple analytic signals, which would also illuminate the size of the prefactor."],"forward_implications":["Observation windows of modestly increased length now provably give exponentially better frequency resolution for analytic quasi-periodic signals, rather than the former power-law improvement.","The exponent zeta makes explicit the trade-off: larger Diophantine exponent tau or less flat weighting functions slow the exponential rate, while flatter weights accelerate it.","The Brjuno extension replaces Diophantine lower bounds with the weakest classical nonresonance condition, giving errors of the form exp(-c log T log log T) even when power-law small divisors are absent.","The almost-periodic extensions show the same convergence mechanism works for infinite frequency sets with weighted spatial structures, connecting the rate to the growth of the spatial weight.","The techniques transfer directly to weighted Birkhoff averages, linking exponential acceleration results in ergodic and numerical analysis to frequency map analysis."],"fun_headline_variants":["Laskar's frequency analysis errors shrink exponentially, new proof","Exponential convergence proven for Laskar's frequency map","Weighting functions make Laskar's method exponentially accurate","First unified exponential theory for frequency map analysis","New proof: Laskar's frequency errors decay faster than any power"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The exponential conclusions hinge on an unproved bound on the L^1 norms of high derivatives of the weighting function w_{p,q}; if that bound fails, the integration-by-parts argument in Lemma 2.3 stops producing exponential decay and the stated rates no longer follow.","fun_headline_variants_meta":{"raw":{"variants":["Laskar's frequency analysis errors shrink exponentially, new proof","Exponential convergence proven for Laskar's frequency map","Weighting functions make Laskar's method exponentially accurate","First unified exponential theory for frequency map analysis","New proof: Laskar's frequency errors decay faster than any power"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00106,"raw_usage":{"total_tokens":4277,"prompt_tokens":729,"completion_tokens":3548,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":473,"completion_tokens_details":{"reasoning_tokens":3466}},"tokens_in":473,"tokens_out":3548,"duration_ms":20136,"temperature":1.0,"reasoning_tokens":3466,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T12:47:34.690001+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a concrete analytic quasi-periodic function with a known Diophantine frequency vector, compute the frequency map estimate nu_T^1 numerically for increasing window lengths T, and check whether |nu_T^1 - nu_1| decays like exp(-c T^zeta) for a range of positive c; alternatively, directly test the bound ||D^l w_{p,q}||_{L^1} <= lambda^l l^{beta l} for large l by symbolic or numerical differentiation, since one counterexample would disable Lemma 2.3.","supporting_citations":[],"review_version":1}